A modular approach for integrative analysis of large-scale gene-expression and drug-response data

High-throughput technologies are now used to generate more than one type of data from the same biological samples. To properly integrate such data, we propose using co-modules, which describe coherent patterns across paired data sets, and conceive several modular methods for their identification. We first test these methods using in silico data, demonstrating that the integrative scheme of our Ping-Pong Algorithm uncovers drug-gene associations more accurately when considering noisy or complex data. Second, we provide an extensive comparative study using the gene-expression and drug-response data from the NCI-60 cell lines. Using information from the DrugBank and the Connectivity Map databases we show that the Ping-Pong Algorithm predicts drug-gene associations significantly better than other methods. Co-modules provide insights into possible mechanisms of action for a wide range of drugs and suggest new targets for therapy

Modular analysis of gene expression data with R.

SUMMARY: Large sets of data, such as expression profiles from many samples, require analytic tools to reduce their complexity. The Iterative Signature Algorithm (ISA) is a biclustering algorithm. It was designed to decompose a large set of data into so-called 'modules'. In the context of gene expression data, these modules consist of subsets of genes that exhibit a coherent expression profile only over a subset of microarray experiments. Genes and arrays may be attributed to multiple modules and the level of required coherence can be varied resulting in different 'resolutions' of the modular mapping. In this short note, we introduce two BioConductor software packages written in GNU R: The isa2 package includes an optimized implementation of the ISA and the eisa package provides a convenient interface to run the ISA, visualize its output and put the biclusters into biological context. Potential users of these packages are all R and BioConductor users dealing with tabular (e.g. gene expression) data. AVAILABILITY: http://www.unil.ch/cbg/ISA CONTACT: sven.bergmann@unil.ch

Identification of corticosteroid-regulated genes in cardiomyocytes by serial analysis of gene expression.

Corticosteroids (aldosterone, cortisol/corticosterone) exert direct functional effects on cardiomyocytes. However, gene networks activated by corticosteroids in cardiomyocytes, as well as the involvement of the mineralocorticoid receptor (MR) vs the glucocorticoid receptor (GR) in these effects, remain largely unknown. Here we characterized the corticosteroid-dependent transcriptome in primary culture of neonatal mouse cardiomyocytes treated with 10(-6) M aldosterone, a concentration predicted to occupy both MR and GR. Serial analysis of gene expression revealed 101 aldosterone-regulated genes. The MR/GR specificity was characterized for one regulated transcript, namely ecto-ADP-ribosyltransferase-3 (Art3). Using cardiomyocytes from GR(null/null) or MR(null/null) mice we demonstrate that in GR(null/null) cardiomyocytes the response is abrogated, but it is fully maintained in MR(null/null) cardiomyocytes. We conclude that Art3 expression is regulated exclusively via the GR. Our study identifies a new set of corticosteroid-regulated genes in cardiomyocytes and demonstrates a new approach to studying the selectivity of MR- vs GR-dependent effects.

Methods for the ex vivo characterization of human CD8+ T subsets based on gene expression and replicative history analysis.

The generation of an antigen-specific T-lymphocyte response is a complex multi-step process. Upon T-cell receptor-mediated recognition of antigen presented by activated dendritic cells, naive T-lymphocytes enter a program of proliferation and differentiation, during the course of which they acquire effector functions and may ultimately become memory T-cells. A major goal of modern immunology is to precisely identify and characterize effector and memory T-cell subpopulations that may be most efficient in disease protection. Sensitive methods are required to address these questions in exceedingly low numbers of antigen-specific lymphocytes recovered from clinical samples, and not manipulated in vitro. We have developed new techniques to dissect immune responses against viral or tumor antigens. These allow the isolation of various subsets of antigen-specific T-cells (with major histocompatibility complex [MHC]-peptide multimers and five-color FACS sorting) and the monitoring of gene expression in individual cells (by five-cell reverse transcription-polymerase chain reaction [RT-PCR]). We can also follow their proliferative life history by flow-fluorescence in situ hybridization (FISH) analysis of average telomere length. Recently, using these tools, we have identified subpopulations of CD8+ T-lymphocytes with distinct proliferative history and partial effector-like properties. Our data suggest that these subsets descend from recently activated T-cells and are committed to become differentiated effector T-lymphocytes.

A comprehensive analysis of gene expression profiles in distal parts of the mouse renal tubule.

The distal parts of the renal tubule play a critical role in maintaining homeostasis of extracellular fluids. In this review, we present an in-depth analysis of microarray-based gene expression profiles available for microdissected mouse distal nephron segments, i.e., the distal convoluted tubule (DCT) and the connecting tubule (CNT), and for the cortical portion of the collecting duct (CCD; Zuber et al., Proc Natl Acad Sci USA 106:16523-16528, 2009). Classification of expressed transcripts in 14 major functional gene categories demonstrated that all principal proteins involved in maintaining the salt and water balance are represented by highly abundant transcripts. However, a significant number of transcripts belonging, for instance, to categories of G-protein-coupled receptors or serine/threonine kinases exhibit high expression levels but remain unassigned to a specific renal function. We also established a list of genes differentially expressed between the DCT/CNT and the CCD. This list is enriched by genes related to segment-specific transport functions and by transcription factors directing the development of the distal nephron or collecting ducts. Collectively, this in silico analysis provides comprehensive information about relative abundance and tissue specificity of the DCT/CNT and the CCD expressed transcripts and identifies new candidate genes for renal homeostasis.

Network-guided analysis of genes with altered somatic copy number and gene expression reveals pathways commonly perturbed in metastatic melanoma.

Cancer genomes frequently contain somatic copy number alterations (SCNA) that can significantly perturb the expression level of affected genes and thus disrupt pathways controlling normal growth. In melanoma, many studies have focussed on the copy number and gene expression levels of the BRAF, PTEN and MITF genes, but little has been done to identify new genes using these parameters at the genome-wide scale. Using karyotyping, SNP and CGH arrays, and RNA-seq, we have identified SCNA affecting gene expression ('SCNA-genes') in seven human metastatic melanoma cell lines. We showed that the combination of these techniques is useful to identify candidate genes potentially involved in tumorigenesis. Since few of these alterations were recurrent across our samples, we used a protein network-guided approach to determine whether any pathways were enriched in SCNA-genes in one or more samples. From this unbiased genome-wide analysis, we identified 28 significantly enriched pathway modules. Comparison with two large, independent melanoma SCNA datasets showed less than 10% overlap at the individual gene level, but network-guided analysis revealed 66% shared pathways, including all but three of the pathways identified in our data. Frequently altered pathways included WNT, cadherin signalling, angiogenesis and melanogenesis. Additionally, our results emphasize the potential of the EPHA3 and FRS2 gene products, involved in angiogenesis and migration, as possible therapeutic targets in melanoma. Our study demonstrates the utility of network-guided approaches, for both large and small datasets, to identify pathways recurrently perturbed in cancer.

Whole genome analysis of p38 SAPK-mediated gene expression upon stress

Background: Cells have the ability to respond and adapt to environmental changes through activation of stress-activated protein kinases (SAPKs). Although p38 SAPK signalling is known to participate in the regulation of gene expression little is known on the molecular mechanisms used by this SAPK to regulate stress-responsive genes and the overall set of genes regulated by p38 in response to different stimuli.Results: Here, we report a whole genome expression analyses on mouse embryonic fibroblasts (MEFs) treated with three different p38 SAPK activating-stimuli, namely osmostress, the cytokine TNFα and the protein synthesis inhibitor anisomycin. We have found that the activation kinetics of p38α SAPK in response to these insults is different and also leads to a complex gene pattern response specific for a given stress with a restricted set of overlapping genes. In addition, we have analysed the contribution of p38α the major p38 family member present in MEFs, to the overall stress-induced transcriptional response by using both a chemical inhibitor (SB203580) and p38α deficient (p38α-/-) MEFs. We show here that p38 SAPK dependency ranged between 60% and 88% depending on the treatments and that there is a very good overlap between the inhibitor treatment and the ko cells. Furthermore, we have found that the dependency of SAPK varies depending on the time the cells are subjected to osmostress. Conclusions: Our genome-wide transcriptional analyses shows a selective response to specific stimuli and a restricted common response of up to 20% of the stress up-regulated early genes that involves an important set of transcription factors, which might be critical for either cell adaptation or preparation for continuous extra-cellular changes. Interestingly, up to 85% of the up-regulated genes are under the transcriptional control of p38 SAPK. Thus, activation of p38 SAPK is critical to elicit the early gene expression program required for cell adaptation to stress.

Differential Gene Expression between Fall- and Spring-Run Chinook Salmon Assessed by Long Serial Analysis of Gene Expression

Of all Pacific salmonids, Chinook salmon Oncorhynchus tshawytscha display the greatest variability in return times to freshwater. The molecular mechanisms of these differential return times have not been well described. Current methods, such as long serial analysis of gene expression (LongSAGE) and microarrays, allow gene expression to be analyzed for thousands of genes simultaneously. To investigate whether differential gene expression is observed between fall- and spring-run Chinook salmon from California's Central Valley, LongSAGE libraries were constructed. Three libraries containing between 25,512 and 29,372 sequenced tags (21 base pairs/tag) were generated using messenger RNA from the brains of adult Chinook salmon returning in fall and spring and from one ocean-caught Chinook salmon. Tags were annotated to genes using complementary DNA libraries from Atlantic salmon Salmo salar and rainbow trout O. mykiss. Differentially expressed genes, as estimated by differences in the number of sequence tags, were found in all pairwise comparisons of libraries (freshwater versus saltwater = 40 genes; fall versus spring = 11 genes: and spawning versus nonspawning = 51 genes). The gene for ependymin, an extracellular glycoprotein involved in behavioral plasticity in fish, exhibited the most differential expression among the three groupings. Reverse transcription polymerase chain reaction analysis verified the differential expression of ependymin between the fall- and spring-run samples. These LongSAGE libraries, the first reported for Chinook salmon, provide a window of the transcriptional changes during Chinook salmon return migration to freshwater and spawning and increase the amount of expressed sequence data.

Integrated Analysis of High-Resolution Gene Expression and Copy Number Profiling Identified Biallelic Deletion of CDKN2A/2B Tumor Suppressor Locus As the Most Frequent and Unique Genomic Abnormality in Diffuse Large B-Cell Lymphoma (DLBCL) with Strong Prognostic Value in Both GCB and ABC Subtypes and Not Overcome by a Dose-Intensive Immunochemotherapy Regimen Plus Rituximab. Results of a Prospective GELA Clinical Trial Program

Background and aim of the study: Genomic gains and losses play a crucial role in the development and progression of DLBCL and are closely related to gene expression profiles (GEP), including the germinal center B-cell like (GCB) and activated B-cell like (ABC) cell of origin (COO) molecular signatures. To identify new oncogenes or tumor suppressor genes (TSG) involved in DLBCL pathogenesis and to determine their prognostic values, an integrated analysis of high-resolution gene expression and copy number profiling was performed. Patients and methods: Two hundred and eight adult patients with de novo CD20+ DLBCL enrolled in the prospective multicentric randomized LNH-03 GELA trials (LNH03-1B, -2B, -3B, 39B, -5B, -6B, -7B) with available frozen tumour samples, centralized reviewing and adequate DNA/RNA quality were selected. 116 patients were treated by Rituximab(R)-CHOP/R-miniCHOP and 92 patients were treated by the high dose (R)-ACVBP regimen dedicated to patients younger than 60 years (y) in frontline. Tumour samples were simultaneously analysed by high resolution comparative genomic hybridization (CGH, Agilent, 144K) and gene expression arrays (Affymetrix, U133+2). Minimal common regions (MCR), as defined by segments that affect the same chromosomal region in different cases, were delineated. Gene expression and MCR data sets were merged using Gene expression and dosage integrator algorithm (GEDI, Lenz et al. PNAS 2008) to identify new potential driver genes. Results: A total of 1363 recurrent (defined by a penetrance > 5%) MCRs within the DLBCL data set, ranging in size from 386 bp, affecting a single gene, to more than 24 Mb were identified by CGH. Of these MCRs, 756 (55%) showed a significant association with gene expression: 396 (59%) gains, 354 (52%) single-copy deletions, and 6 (67%) homozygous deletions. By this integrated approach, in addition to previously reported genes (CDKN2A/2B, PTEN, DLEU2, TNFAIP3, B2M, CD58, TNFRSF14, FOXP1, REL...), several genes targeted by gene copy abnormalities with a dosage effect and potential physiopathological impact were identified, including genes with TSG activity involved in cell cycle (HACE1, CDKN2C) immune response (CD68, CD177, CD70, TNFSF9, IRAK2), DNA integrity (XRCC2, BRCA1, NCOR1, NF1, FHIT) or oncogenic functions (CD79b, PTPRT, MALT1, AUTS2, MCL1, PTTG1...) with distinct distribution according to COO signature. The CDKN2A/2B tumor suppressor locus (9p21) was deleted homozygously in 27% of cases and hemizygously in 9% of cases. Biallelic loss was observed in 49% of ABC DLBCL and in 10% of GCB DLBCL. This deletion was strongly correlated to age and associated to a limited number of additional genetic abnormalities including trisomy 3, 18 and short gains/losses of Chr. 1, 2, 19 regions (FDR < 0.01), allowing to identify genes that may have synergistic effects with CDKN2A/2B inactivation. With a median follow-up of 42.9 months, only CDKN2A/2B biallelic deletion strongly correlates (FDR p.value < 0.01) to a poor outcome in the entire cohort (4y PFS = 44% [32-61] respectively vs. 74% [66-82] for patients in germline configuration; 4y OS = 53% [39-72] vs 83% [76-90]). In a Cox proportional hazard prediction of the PFS, CDKN2A/2B deletion remains predictive (HR = 1.9 [1.1-3.2], p = 0.02) when combined with IPI (HR = 2.4 [1.4-4.1], p = 0.001) and GCB status (HR = 1.3 [0.8-2.3], p = 0.31). This difference remains predictive in the subgroup of patients treated by R-CHOP (4y PFS = 43% [29-63] vs. 66% [55-78], p=0.02), in patients treated by R-ACVBP (4y PFS = 49% [28-84] vs. 83% [74-92], p=0.003), and in GCB (4y PFS = 50% [27-93] vs. 81% [73-90], p=0.02), or ABC/unclassified (5y PFS = 42% [28-61] vs. 67% [55-82] p = 0.009) molecular subtypes (Figure 1). Conclusion: We report for the first time an integrated genetic analysis of a large cohort of DLBCL patients included in a prospective multicentric clinical trial program allowing identifying new potential driver genes with pathogenic impact. However CDKN2A/2B deletion constitutes the strongest and unique prognostic factor of chemoresistance to R-CHOP, regardless the COO signature, which is not overcome by a more intensified immunochemotherapy. Patients displaying this frequent genomic abnormality warrant new and dedicated therapeutic approaches.

Comparative modular analysis of gene expression in vertebrate organs.

ABSTRACT: BACKGROUND: The degree of conservation of gene expression between homologous organs largely remains an open question. Several recent studies reported some evidence in favor of such conservation. Most studies compute organs' similarity across all orthologous genes, whereas the expression level of many genes are not informative about organ specificity. RESULTS: Here, we use a modularization algorithm to overcome this limitation through the identification of inter-species co-modules of organs and genes. We identify such co-modules using mouse and human microarray expression data. They are functionally coherent both in terms of genes and of organs from both organisms. We show that a large proportion of genes belonging to the same co-module are orthologous between mouse and human. Moreover, their zebrafish orthologs also tend to be expressed in the corresponding homologous organs. Notable exceptions to the general pattern of conservation are the testis and the olfactory bulb. Interestingly, some co-modules consist of single organs, while others combine several functionally related organs. For instance, amygdala, cerebral cortex, hypothalamus and spinal cord form a clearly discernible unit of expression, both in mouse and human. CONCLUSIONS: Our study provides a new framework for comparative analysis which will be applicable also to other sets of large-scale phenotypic data collected across different species.

Analysis of gene expression patterns in animals

During my PhD, my aim was to provide new tools to increase our capacity to analyse gene expression patterns, and to study on a large-scale basis the evolution of gene expression in animals. Gene expression patterns (when and where a gene is expressed) are a key feature in understanding gene function, notably in development. It appears clear now that the evolution of developmental processes and of phenotypes is shaped both by evolution at the coding sequence level, and at the gene expression level.Studying gene expression evolution in animals, with complex expression patterns over tissues and developmental time, is still challenging. No tools are available to routinely compare expression patterns between different species, with precision, and on a large-scale basis. Studies on gene expression evolution are therefore performed only on small genes datasets, or using imprecise descriptions of expression patterns.The aim of my PhD was thus to develop and use novel bioinformatics resources, to study the evolution of gene expression. To this end, I developed the database Bgee (Base for Gene Expression Evolution). The approach of Bgee is to transform heterogeneous expression data (ESTs, microarrays, and in-situ hybridizations) into present/absent calls, and to annotate them to standard representations of anatomy and development of different species (anatomical ontologies). An extensive mapping between anatomies of species is then developed based on hypothesis of homology. These precise annotations to anatomies, and this extensive mapping between species, are the major assets of Bgee, and have required the involvement of many co-workers over the years. My main personal contribution is the development and the management of both the Bgee database and the web-application.Bgee is now on its ninth release, and includes an important gene expression dataset for 5 species (human, mouse, drosophila, zebrafish, Xenopus), with the most data from mouse, human and zebrafish. Using these three species, I have conducted an analysis of gene expression evolution after duplication in vertebrates.Gene duplication is thought to be a major source of novelty in evolution, and to participate to speciation. It has been suggested that the evolution of gene expression patterns might participate in the retention of duplicate genes. I performed a large-scale comparison of expression patterns of hundreds of duplicated genes to their singleton ortholog in an outgroup, including both small and large-scale duplicates, in three vertebrate species (human, mouse and zebrafish), and using highly accurate descriptions of expression patterns. My results showed unexpectedly high rates of de novo acquisition of expression domains after duplication (neofunctionalization), at least as high or higher than rates of partitioning of expression domains (subfunctionalization). I found differences in the evolution of expression of small- and large-scale duplicates, with small-scale duplicates more prone to neofunctionalization. Duplicates with neofunctionalization seemed to evolve under more relaxed selective pressure on the coding sequence. Finally, even with abundant and precise expression data, the majority fate I recovered was neither neo- nor subfunctionalization of expression domains, suggesting a major role for other mechanisms in duplicate gene retention.

Meta-analysis of gene expression profiles in breast cancer: toward a unified understanding of breast cancer subtyping and prognosis signatures.

INTRODUCTION: Breast cancer subtyping and prognosis have been studied extensively by gene expression profiling, resulting in disparate signatures with little overlap in their constituent genes. Although a previous study demonstrated a prognostic concordance among gene expression signatures, it was limited to only one dataset and did not fully elucidate how the different genes were related to one another nor did it examine the contribution of well-known biological processes of breast cancer tumorigenesis to their prognostic performance. METHOD: To address the above issues and to further validate these initial findings, we performed the largest meta-analysis of publicly available breast cancer gene expression and clinical data, which are comprised of 2,833 breast tumors. Gene coexpression modules of three key biological processes in breast cancer (namely, proliferation, estrogen receptor [ER], and HER2 signaling) were used to dissect the role of constituent genes of nine prognostic signatures. RESULTS: Using a meta-analytical approach, we consolidated the signatures associated with ER signaling, ERBB2 amplification, and proliferation. Previously published expression-based nomenclature of breast cancer 'intrinsic' subtypes can be mapped to the three modules, namely, the ER-/HER2- (basal-like), the HER2+ (HER2-like), and the low- and high-proliferation ER+/HER2- subtypes (luminal A and B). We showed that all nine prognostic signatures exhibited a similar prognostic performance in the entire dataset. Their prognostic abilities are due mostly to the detection of proliferation activity. Although ER- status (basal-like) and ERBB2+ expression status correspond to bad outcome, they seem to act through elevated expression of proliferation genes and thus contain only indirect information about prognosis. Clinical variables measuring the extent of tumor progression, such as tumor size and nodal status, still add independent prognostic information to proliferation genes. CONCLUSION: This meta-analysis unifies various results of previous gene expression studies in breast cancer. It reveals connections between traditional prognostic factors, expression-based subtyping, and prognostic signatures, highlighting the important role of proliferation in breast cancer prognosis.

A comprehensive analysis of human gene expression profiles identifies stromal immunoglobulin κ C as a compatible prognostic marker in human solid tumors.

PURPOSE: Although the central role of the immune system for tumor prognosis is generally accepted, a single robust marker is not yet available. EXPERIMENTAL DESIGN: On the basis of receiver operating characteristic analyses, robust markers were identified from a 60-gene B cell-derived metagene and analyzed in gene expression profiles of 1,810 breast cancer; 1,056 non-small cell lung carcinoma (NSCLC); 513 colorectal; and 426 ovarian cancer patients. Protein and RNA levels were examined in paraffin-embedded tissue of 330 breast cancer patients. The cell types were identified with immunohistochemical costaining and confocal fluorescence microscopy. RESULTS: We identified immunoglobulin κ C (IGKC) which as a single marker is similarly predictive and prognostic as the entire B-cell metagene. IGKC was consistently associated with metastasis-free survival across different molecular subtypes in node-negative breast cancer (n = 965) and predicted response to anthracycline-based neoadjuvant chemotherapy (n = 845; P < 0.001). In addition, IGKC gene expression was prognostic in NSCLC and colorectal cancer. No association was observed in ovarian cancer. IGKC protein expression was significantly associated with survival in paraffin-embedded tissues of 330 breast cancer patients. Tumor-infiltrating plasma cells were identified as the source of IGKC expression. CONCLUSION: Our findings provide IGKC as a novel diagnostic marker for risk stratification in human cancer and support concepts to exploit the humoral immune response for anticancer therapy. It could be validated in several independent cohorts and carried out similarly well in RNA from fresh frozen as well as from paraffin tissue and on protein level by immunostaining.

Comparative modular analysis of gene expression in vertebrate development

The focus of my PhD research was the concept of modularity. In the last 15 years, modularity has become a classic term in different fields of biology. On the conceptual level, a module is a set of interacting elements that remain mostly independent from the elements outside of the module. I used modular analysis techniques to study gene expression evolution in vertebrates. In particular, I identified ``natural'' modules of gene expression in mouse and human, and I showed that expression of organ-specific and system-specific genes tends to be conserved between such distance vertebrates as mammals and fishes. Also with a modular approach, I studied patterns of developmental constraints on transcriptome evolution. I showed that none of the two commonly accepted models of the evolution of embryonic development (``evo-devo'') are exclusively valid. In particular, I found that the conservation of the sequences of regulatory regions is highest during mid-development of zebrafish, and thus it supports the ``hourglass model''. In contrast, events of gene duplication and new gene introduction are most rare in early development, which supports the ``early conservation model''. In addition to the biological insights on transcriptome evolution, I have also discussed in detail the advantages of modular approaches in large-scale data analysis. Moreover, I re-analyzed several studies (published in high-ranking journals), and showed that their conclusions do not hold out under a detailed analysis. This demonstrates that complex analysis of high-throughput data requires a co-operation between biologists, bioinformaticians, and statisticians.

Gene expression profiling analysis of lung adenocarcinoma

The present study screened potential genes related to lung adenocarcinoma, with the aim of further understanding disease pathogenesis. The GSE2514 dataset including 20 lung adenocarcinoma and 19 adjacent normal tissue samples from 10 patients with lung adenocarcinoma aged 45-73 years was downloaded from Gene Expression Omnibus. Differentially expressed genes (DEGs) between the two groups were screened using the t-test. Potential gene functions were predicted using functional and pathway enrichment analysis, and protein-protein interaction (PPI) networks obtained from the STRING database were constructed with Cytoscape. Module analysis of PPI networks was performed through MCODE in Cytoscape. In total, 535 upregulated and 465 downregulated DEGs were identified. These included ATP5D, UQCRC2, UQCR11 and genes encoding nicotinamide adenine dinucleotide (NADH), which are mainly associated with mitochondrial ATP synthesis coupled electron transport, and which were enriched in the oxidative phosphorylation pathway. Other DEGs were associated with DNA replication (PRIM1, MCM3, and RNASEH2A), cell surface receptor-linked signal transduction and the enzyme-linked receptor protein signaling pathway (MAPK1, STAT3, RAF1, and JAK1), and regulation of the cytoskeleton and phosphatidylinositol signaling system (PIP5K1B, PIP5K1C, and PIP4K2B). Our findings suggest that DEGs encoding subunits of NADH, PRIM1, MCM3, MAPK1, STAT3, RAF1, and JAK1 might be associated with the development of lung adenocarcinoma.